xref: /freebsd/sys/powerpc/booke/pmap.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
1 /*-
2  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
3  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
18  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
19  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
20  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
21  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
22  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
23  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
24  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  * Some hw specific parts of this pmap were derived or influenced
27  * by NetBSD's ibm4xx pmap module. More generic code is shared with
28  * a few other pmap modules from the FreeBSD tree.
29  */
30 
31  /*
32   * VM layout notes:
33   *
34   * Kernel and user threads run within one common virtual address space
35   * defined by AS=0.
36   *
37   * Virtual address space layout:
38   * -----------------------------
39   * 0x0000_0000 - 0xafff_ffff	: user process
40   * 0xb000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
41   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
42   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
43   * 0xc100_0000 - 0xfeef_ffff	: KVA
44   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
45   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
46   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
47   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
48   * 0xfef0_0000 - 0xffff_ffff	: I/O devices region
49   */
50 
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
53 
54 #include <sys/types.h>
55 #include <sys/param.h>
56 #include <sys/malloc.h>
57 #include <sys/ktr.h>
58 #include <sys/proc.h>
59 #include <sys/user.h>
60 #include <sys/queue.h>
61 #include <sys/systm.h>
62 #include <sys/kernel.h>
63 #include <sys/msgbuf.h>
64 #include <sys/lock.h>
65 #include <sys/mutex.h>
66 #include <sys/smp.h>
67 #include <sys/vmmeter.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_page.h>
71 #include <vm/vm_kern.h>
72 #include <vm/vm_pageout.h>
73 #include <vm/vm_extern.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_param.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_pager.h>
78 #include <vm/uma.h>
79 
80 #include <machine/cpu.h>
81 #include <machine/pcb.h>
82 #include <machine/platform.h>
83 
84 #include <machine/tlb.h>
85 #include <machine/spr.h>
86 #include <machine/vmparam.h>
87 #include <machine/md_var.h>
88 #include <machine/mmuvar.h>
89 #include <machine/pmap.h>
90 #include <machine/pte.h>
91 
92 #include "mmu_if.h"
93 
94 #ifdef  DEBUG
95 #define debugf(fmt, args...) printf(fmt, ##args)
96 #else
97 #define debugf(fmt, args...)
98 #endif
99 
100 #define TODO			panic("%s: not implemented", __func__);
101 
102 #include "opt_sched.h"
103 #ifndef SCHED_4BSD
104 #error "e500 only works with SCHED_4BSD which uses a global scheduler lock."
105 #endif
106 extern struct mtx sched_lock;
107 
108 extern int dumpsys_minidump;
109 
110 extern unsigned char _etext[];
111 extern unsigned char _end[];
112 
113 /* Kernel physical load address. */
114 extern uint32_t kernload;
115 vm_offset_t kernstart;
116 vm_size_t kernsize;
117 
118 /* Message buffer and tables. */
119 static vm_offset_t data_start;
120 static vm_size_t data_end;
121 
122 /* Phys/avail memory regions. */
123 static struct mem_region *availmem_regions;
124 static int availmem_regions_sz;
125 static struct mem_region *physmem_regions;
126 static int physmem_regions_sz;
127 
128 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
129 static vm_offset_t zero_page_va;
130 static struct mtx zero_page_mutex;
131 
132 static struct mtx tlbivax_mutex;
133 
134 /*
135  * Reserved KVA space for mmu_booke_zero_page_idle. This is used
136  * by idle thred only, no lock required.
137  */
138 static vm_offset_t zero_page_idle_va;
139 
140 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
141 static vm_offset_t copy_page_src_va;
142 static vm_offset_t copy_page_dst_va;
143 static struct mtx copy_page_mutex;
144 
145 /**************************************************************************/
146 /* PMAP */
147 /**************************************************************************/
148 
149 static void mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
150     vm_prot_t, boolean_t);
151 
152 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
153 unsigned int kernel_ptbls;	/* Number of KVA ptbls. */
154 
155 /*
156  * If user pmap is processed with mmu_booke_remove and the resident count
157  * drops to 0, there are no more pages to remove, so we need not continue.
158  */
159 #define PMAP_REMOVE_DONE(pmap) \
160 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
161 
162 extern void tid_flush(tlbtid_t);
163 
164 /**************************************************************************/
165 /* TLB and TID handling */
166 /**************************************************************************/
167 
168 /* Translation ID busy table */
169 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
170 
171 /*
172  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
173  * core revisions and should be read from h/w registers during early config.
174  */
175 uint32_t tlb0_entries;
176 uint32_t tlb0_ways;
177 uint32_t tlb0_entries_per_way;
178 
179 #define TLB0_ENTRIES		(tlb0_entries)
180 #define TLB0_WAYS		(tlb0_ways)
181 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
182 
183 #define TLB1_ENTRIES 16
184 
185 /* In-ram copy of the TLB1 */
186 static tlb_entry_t tlb1[TLB1_ENTRIES];
187 
188 /* Next free entry in the TLB1 */
189 static unsigned int tlb1_idx;
190 
191 static tlbtid_t tid_alloc(struct pmap *);
192 
193 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
194 
195 static int tlb1_set_entry(vm_offset_t, vm_offset_t, vm_size_t, uint32_t);
196 static void tlb1_write_entry(unsigned int);
197 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
198 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_offset_t, vm_size_t);
199 
200 static vm_size_t tsize2size(unsigned int);
201 static unsigned int size2tsize(vm_size_t);
202 static unsigned int ilog2(unsigned int);
203 
204 static void set_mas4_defaults(void);
205 
206 static inline void tlb0_flush_entry(vm_offset_t);
207 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
208 
209 /**************************************************************************/
210 /* Page table management */
211 /**************************************************************************/
212 
213 /* Data for the pv entry allocation mechanism */
214 static uma_zone_t pvzone;
215 static struct vm_object pvzone_obj;
216 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
217 
218 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
219 
220 #ifndef PMAP_SHPGPERPROC
221 #define PMAP_SHPGPERPROC	200
222 #endif
223 
224 static void ptbl_init(void);
225 static struct ptbl_buf *ptbl_buf_alloc(void);
226 static void ptbl_buf_free(struct ptbl_buf *);
227 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
228 
229 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int);
230 static void ptbl_free(mmu_t, pmap_t, unsigned int);
231 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
232 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
233 
234 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
235 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
236 static void pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t);
237 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
238 
239 static pv_entry_t pv_alloc(void);
240 static void pv_free(pv_entry_t);
241 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
242 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
243 
244 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
245 #define PTBL_BUFS		(128 * 16)
246 
247 struct ptbl_buf {
248 	TAILQ_ENTRY(ptbl_buf) link;	/* list link */
249 	vm_offset_t kva;		/* va of mapping */
250 };
251 
252 /* ptbl free list and a lock used for access synchronization. */
253 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
254 static struct mtx ptbl_buf_freelist_lock;
255 
256 /* Base address of kva space allocated fot ptbl bufs. */
257 static vm_offset_t ptbl_buf_pool_vabase;
258 
259 /* Pointer to ptbl_buf structures. */
260 static struct ptbl_buf *ptbl_bufs;
261 
262 void pmap_bootstrap_ap(volatile uint32_t *);
263 
264 /*
265  * Kernel MMU interface
266  */
267 static void		mmu_booke_change_wiring(mmu_t, pmap_t, vm_offset_t, boolean_t);
268 static void		mmu_booke_clear_modify(mmu_t, vm_page_t);
269 static void		mmu_booke_clear_reference(mmu_t, vm_page_t);
270 static void		mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
271     vm_size_t, vm_offset_t);
272 static void		mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
273 static void		mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
274     vm_prot_t, boolean_t);
275 static void		mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
276     vm_page_t, vm_prot_t);
277 static void		mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
278     vm_prot_t);
279 static vm_paddr_t	mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
280 static vm_page_t	mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
281     vm_prot_t);
282 static void		mmu_booke_init(mmu_t);
283 static boolean_t	mmu_booke_is_modified(mmu_t, vm_page_t);
284 static boolean_t	mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
285 static boolean_t	mmu_booke_is_referenced(mmu_t, vm_page_t);
286 static boolean_t	mmu_booke_ts_referenced(mmu_t, vm_page_t);
287 static vm_offset_t	mmu_booke_map(mmu_t, vm_offset_t *, vm_offset_t, vm_offset_t,
288     int);
289 static int		mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
290     vm_paddr_t *);
291 static void		mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
292     vm_object_t, vm_pindex_t, vm_size_t);
293 static boolean_t	mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
294 static void		mmu_booke_page_init(mmu_t, vm_page_t);
295 static int		mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
296 static void		mmu_booke_pinit(mmu_t, pmap_t);
297 static void		mmu_booke_pinit0(mmu_t, pmap_t);
298 static void		mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
299     vm_prot_t);
300 static void		mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
301 static void		mmu_booke_qremove(mmu_t, vm_offset_t, int);
302 static void		mmu_booke_release(mmu_t, pmap_t);
303 static void		mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
304 static void		mmu_booke_remove_all(mmu_t, vm_page_t);
305 static void		mmu_booke_remove_write(mmu_t, vm_page_t);
306 static void		mmu_booke_zero_page(mmu_t, vm_page_t);
307 static void		mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
308 static void		mmu_booke_zero_page_idle(mmu_t, vm_page_t);
309 static void		mmu_booke_activate(mmu_t, struct thread *);
310 static void		mmu_booke_deactivate(mmu_t, struct thread *);
311 static void		mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
312 static void		*mmu_booke_mapdev(mmu_t, vm_offset_t, vm_size_t);
313 static void		mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
314 static vm_offset_t	mmu_booke_kextract(mmu_t, vm_offset_t);
315 static void		mmu_booke_kenter(mmu_t, vm_offset_t, vm_offset_t);
316 static void		mmu_booke_kremove(mmu_t, vm_offset_t);
317 static boolean_t	mmu_booke_dev_direct_mapped(mmu_t, vm_offset_t, vm_size_t);
318 static void		mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
319     vm_size_t);
320 static vm_offset_t	mmu_booke_dumpsys_map(mmu_t, struct pmap_md *,
321     vm_size_t, vm_size_t *);
322 static void		mmu_booke_dumpsys_unmap(mmu_t, struct pmap_md *,
323     vm_size_t, vm_offset_t);
324 static struct pmap_md	*mmu_booke_scan_md(mmu_t, struct pmap_md *);
325 
326 static mmu_method_t mmu_booke_methods[] = {
327 	/* pmap dispatcher interface */
328 	MMUMETHOD(mmu_change_wiring,	mmu_booke_change_wiring),
329 	MMUMETHOD(mmu_clear_modify,	mmu_booke_clear_modify),
330 	MMUMETHOD(mmu_clear_reference,	mmu_booke_clear_reference),
331 	MMUMETHOD(mmu_copy,		mmu_booke_copy),
332 	MMUMETHOD(mmu_copy_page,	mmu_booke_copy_page),
333 	MMUMETHOD(mmu_enter,		mmu_booke_enter),
334 	MMUMETHOD(mmu_enter_object,	mmu_booke_enter_object),
335 	MMUMETHOD(mmu_enter_quick,	mmu_booke_enter_quick),
336 	MMUMETHOD(mmu_extract,		mmu_booke_extract),
337 	MMUMETHOD(mmu_extract_and_hold,	mmu_booke_extract_and_hold),
338 	MMUMETHOD(mmu_init,		mmu_booke_init),
339 	MMUMETHOD(mmu_is_modified,	mmu_booke_is_modified),
340 	MMUMETHOD(mmu_is_prefaultable,	mmu_booke_is_prefaultable),
341 	MMUMETHOD(mmu_is_referenced,	mmu_booke_is_referenced),
342 	MMUMETHOD(mmu_ts_referenced,	mmu_booke_ts_referenced),
343 	MMUMETHOD(mmu_map,		mmu_booke_map),
344 	MMUMETHOD(mmu_mincore,		mmu_booke_mincore),
345 	MMUMETHOD(mmu_object_init_pt,	mmu_booke_object_init_pt),
346 	MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
347 	MMUMETHOD(mmu_page_init,	mmu_booke_page_init),
348 	MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
349 	MMUMETHOD(mmu_pinit,		mmu_booke_pinit),
350 	MMUMETHOD(mmu_pinit0,		mmu_booke_pinit0),
351 	MMUMETHOD(mmu_protect,		mmu_booke_protect),
352 	MMUMETHOD(mmu_qenter,		mmu_booke_qenter),
353 	MMUMETHOD(mmu_qremove,		mmu_booke_qremove),
354 	MMUMETHOD(mmu_release,		mmu_booke_release),
355 	MMUMETHOD(mmu_remove,		mmu_booke_remove),
356 	MMUMETHOD(mmu_remove_all,	mmu_booke_remove_all),
357 	MMUMETHOD(mmu_remove_write,	mmu_booke_remove_write),
358 	MMUMETHOD(mmu_sync_icache,	mmu_booke_sync_icache),
359 	MMUMETHOD(mmu_zero_page,	mmu_booke_zero_page),
360 	MMUMETHOD(mmu_zero_page_area,	mmu_booke_zero_page_area),
361 	MMUMETHOD(mmu_zero_page_idle,	mmu_booke_zero_page_idle),
362 	MMUMETHOD(mmu_activate,		mmu_booke_activate),
363 	MMUMETHOD(mmu_deactivate,	mmu_booke_deactivate),
364 
365 	/* Internal interfaces */
366 	MMUMETHOD(mmu_bootstrap,	mmu_booke_bootstrap),
367 	MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
368 	MMUMETHOD(mmu_mapdev,		mmu_booke_mapdev),
369 	MMUMETHOD(mmu_kenter,		mmu_booke_kenter),
370 	MMUMETHOD(mmu_kextract,		mmu_booke_kextract),
371 /*	MMUMETHOD(mmu_kremove,		mmu_booke_kremove),	*/
372 	MMUMETHOD(mmu_unmapdev,		mmu_booke_unmapdev),
373 
374 	/* dumpsys() support */
375 	MMUMETHOD(mmu_dumpsys_map,	mmu_booke_dumpsys_map),
376 	MMUMETHOD(mmu_dumpsys_unmap,	mmu_booke_dumpsys_unmap),
377 	MMUMETHOD(mmu_scan_md,		mmu_booke_scan_md),
378 
379 	{ 0, 0 }
380 };
381 
382 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
383 
384 static inline void
385 tlb_miss_lock(void)
386 {
387 #ifdef SMP
388 	struct pcpu *pc;
389 
390 	if (!smp_started)
391 		return;
392 
393 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
394 		if (pc != pcpup) {
395 
396 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
397 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
398 
399 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
400 			    ("tlb_miss_lock: tried to lock self"));
401 
402 			tlb_lock(pc->pc_booke_tlb_lock);
403 
404 			CTR1(KTR_PMAP, "%s: locked", __func__);
405 		}
406 	}
407 #endif
408 }
409 
410 static inline void
411 tlb_miss_unlock(void)
412 {
413 #ifdef SMP
414 	struct pcpu *pc;
415 
416 	if (!smp_started)
417 		return;
418 
419 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
420 		if (pc != pcpup) {
421 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
422 			    __func__, pc->pc_cpuid);
423 
424 			tlb_unlock(pc->pc_booke_tlb_lock);
425 
426 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
427 		}
428 	}
429 #endif
430 }
431 
432 /* Return number of entries in TLB0. */
433 static __inline void
434 tlb0_get_tlbconf(void)
435 {
436 	uint32_t tlb0_cfg;
437 
438 	tlb0_cfg = mfspr(SPR_TLB0CFG);
439 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
440 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
441 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
442 }
443 
444 /* Initialize pool of kva ptbl buffers. */
445 static void
446 ptbl_init(void)
447 {
448 	int i;
449 
450 	CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
451 	    (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
452 	CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
453 	    __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
454 
455 	mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
456 	TAILQ_INIT(&ptbl_buf_freelist);
457 
458 	for (i = 0; i < PTBL_BUFS; i++) {
459 		ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
460 		TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
461 	}
462 }
463 
464 /* Get a ptbl_buf from the freelist. */
465 static struct ptbl_buf *
466 ptbl_buf_alloc(void)
467 {
468 	struct ptbl_buf *buf;
469 
470 	mtx_lock(&ptbl_buf_freelist_lock);
471 	buf = TAILQ_FIRST(&ptbl_buf_freelist);
472 	if (buf != NULL)
473 		TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
474 	mtx_unlock(&ptbl_buf_freelist_lock);
475 
476 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
477 
478 	return (buf);
479 }
480 
481 /* Return ptbl buff to free pool. */
482 static void
483 ptbl_buf_free(struct ptbl_buf *buf)
484 {
485 
486 	CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
487 
488 	mtx_lock(&ptbl_buf_freelist_lock);
489 	TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
490 	mtx_unlock(&ptbl_buf_freelist_lock);
491 }
492 
493 /*
494  * Search the list of allocated ptbl bufs and find on list of allocated ptbls
495  */
496 static void
497 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
498 {
499 	struct ptbl_buf *pbuf;
500 
501 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
502 
503 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
504 
505 	TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
506 		if (pbuf->kva == (vm_offset_t)ptbl) {
507 			/* Remove from pmap ptbl buf list. */
508 			TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
509 
510 			/* Free corresponding ptbl buf. */
511 			ptbl_buf_free(pbuf);
512 			break;
513 		}
514 }
515 
516 /* Allocate page table. */
517 static pte_t *
518 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
519 {
520 	vm_page_t mtbl[PTBL_PAGES];
521 	vm_page_t m;
522 	struct ptbl_buf *pbuf;
523 	unsigned int pidx;
524 	pte_t *ptbl;
525 	int i;
526 
527 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
528 	    (pmap == kernel_pmap), pdir_idx);
529 
530 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
531 	    ("ptbl_alloc: invalid pdir_idx"));
532 	KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
533 	    ("pte_alloc: valid ptbl entry exists!"));
534 
535 	pbuf = ptbl_buf_alloc();
536 	if (pbuf == NULL)
537 		panic("pte_alloc: couldn't alloc kernel virtual memory");
538 
539 	ptbl = (pte_t *)pbuf->kva;
540 
541 	CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
542 
543 	/* Allocate ptbl pages, this will sleep! */
544 	for (i = 0; i < PTBL_PAGES; i++) {
545 		pidx = (PTBL_PAGES * pdir_idx) + i;
546 		while ((m = vm_page_alloc(NULL, pidx,
547 		    VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
548 
549 			PMAP_UNLOCK(pmap);
550 			vm_page_unlock_queues();
551 			VM_WAIT;
552 			vm_page_lock_queues();
553 			PMAP_LOCK(pmap);
554 		}
555 		mtbl[i] = m;
556 	}
557 
558 	/* Map allocated pages into kernel_pmap. */
559 	mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
560 
561 	/* Zero whole ptbl. */
562 	bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
563 
564 	/* Add pbuf to the pmap ptbl bufs list. */
565 	TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
566 
567 	return (ptbl);
568 }
569 
570 /* Free ptbl pages and invalidate pdir entry. */
571 static void
572 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
573 {
574 	pte_t *ptbl;
575 	vm_paddr_t pa;
576 	vm_offset_t va;
577 	vm_page_t m;
578 	int i;
579 
580 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
581 	    (pmap == kernel_pmap), pdir_idx);
582 
583 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
584 	    ("ptbl_free: invalid pdir_idx"));
585 
586 	ptbl = pmap->pm_pdir[pdir_idx];
587 
588 	CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
589 
590 	KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
591 
592 	/*
593 	 * Invalidate the pdir entry as soon as possible, so that other CPUs
594 	 * don't attempt to look up the page tables we are releasing.
595 	 */
596 	mtx_lock_spin(&tlbivax_mutex);
597 	tlb_miss_lock();
598 
599 	pmap->pm_pdir[pdir_idx] = NULL;
600 
601 	tlb_miss_unlock();
602 	mtx_unlock_spin(&tlbivax_mutex);
603 
604 	for (i = 0; i < PTBL_PAGES; i++) {
605 		va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
606 		pa = pte_vatopa(mmu, kernel_pmap, va);
607 		m = PHYS_TO_VM_PAGE(pa);
608 		vm_page_free_zero(m);
609 		atomic_subtract_int(&cnt.v_wire_count, 1);
610 		mmu_booke_kremove(mmu, va);
611 	}
612 
613 	ptbl_free_pmap_ptbl(pmap, ptbl);
614 }
615 
616 /*
617  * Decrement ptbl pages hold count and attempt to free ptbl pages.
618  * Called when removing pte entry from ptbl.
619  *
620  * Return 1 if ptbl pages were freed.
621  */
622 static int
623 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
624 {
625 	pte_t *ptbl;
626 	vm_paddr_t pa;
627 	vm_page_t m;
628 	int i;
629 
630 	CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
631 	    (pmap == kernel_pmap), pdir_idx);
632 
633 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
634 	    ("ptbl_unhold: invalid pdir_idx"));
635 	KASSERT((pmap != kernel_pmap),
636 	    ("ptbl_unhold: unholding kernel ptbl!"));
637 
638 	ptbl = pmap->pm_pdir[pdir_idx];
639 
640 	//debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
641 	KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
642 	    ("ptbl_unhold: non kva ptbl"));
643 
644 	/* decrement hold count */
645 	for (i = 0; i < PTBL_PAGES; i++) {
646 		pa = pte_vatopa(mmu, kernel_pmap,
647 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
648 		m = PHYS_TO_VM_PAGE(pa);
649 		m->wire_count--;
650 	}
651 
652 	/*
653 	 * Free ptbl pages if there are no pte etries in this ptbl.
654 	 * wire_count has the same value for all ptbl pages, so check the last
655 	 * page.
656 	 */
657 	if (m->wire_count == 0) {
658 		ptbl_free(mmu, pmap, pdir_idx);
659 
660 		//debugf("ptbl_unhold: e (freed ptbl)\n");
661 		return (1);
662 	}
663 
664 	return (0);
665 }
666 
667 /*
668  * Increment hold count for ptbl pages. This routine is used when a new pte
669  * entry is being inserted into the ptbl.
670  */
671 static void
672 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
673 {
674 	vm_paddr_t pa;
675 	pte_t *ptbl;
676 	vm_page_t m;
677 	int i;
678 
679 	CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
680 	    pdir_idx);
681 
682 	KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
683 	    ("ptbl_hold: invalid pdir_idx"));
684 	KASSERT((pmap != kernel_pmap),
685 	    ("ptbl_hold: holding kernel ptbl!"));
686 
687 	ptbl = pmap->pm_pdir[pdir_idx];
688 
689 	KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
690 
691 	for (i = 0; i < PTBL_PAGES; i++) {
692 		pa = pte_vatopa(mmu, kernel_pmap,
693 		    (vm_offset_t)ptbl + (i * PAGE_SIZE));
694 		m = PHYS_TO_VM_PAGE(pa);
695 		m->wire_count++;
696 	}
697 }
698 
699 /* Allocate pv_entry structure. */
700 pv_entry_t
701 pv_alloc(void)
702 {
703 	pv_entry_t pv;
704 
705 	pv_entry_count++;
706 	if (pv_entry_count > pv_entry_high_water)
707 		pagedaemon_wakeup();
708 	pv = uma_zalloc(pvzone, M_NOWAIT);
709 
710 	return (pv);
711 }
712 
713 /* Free pv_entry structure. */
714 static __inline void
715 pv_free(pv_entry_t pve)
716 {
717 
718 	pv_entry_count--;
719 	uma_zfree(pvzone, pve);
720 }
721 
722 
723 /* Allocate and initialize pv_entry structure. */
724 static void
725 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
726 {
727 	pv_entry_t pve;
728 
729 	//int su = (pmap == kernel_pmap);
730 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
731 	//	(u_int32_t)pmap, va, (u_int32_t)m);
732 
733 	pve = pv_alloc();
734 	if (pve == NULL)
735 		panic("pv_insert: no pv entries!");
736 
737 	pve->pv_pmap = pmap;
738 	pve->pv_va = va;
739 
740 	/* add to pv_list */
741 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
742 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
743 
744 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
745 
746 	//debugf("pv_insert: e\n");
747 }
748 
749 /* Destroy pv entry. */
750 static void
751 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
752 {
753 	pv_entry_t pve;
754 
755 	//int su = (pmap == kernel_pmap);
756 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
757 
758 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
759 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
760 
761 	/* find pv entry */
762 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
763 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
764 			/* remove from pv_list */
765 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
766 			if (TAILQ_EMPTY(&m->md.pv_list))
767 				vm_page_flag_clear(m, PG_WRITEABLE);
768 
769 			/* free pv entry struct */
770 			pv_free(pve);
771 			break;
772 		}
773 	}
774 
775 	//debugf("pv_remove: e\n");
776 }
777 
778 /*
779  * Clean pte entry, try to free page table page if requested.
780  *
781  * Return 1 if ptbl pages were freed, otherwise return 0.
782  */
783 static int
784 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
785 {
786 	unsigned int pdir_idx = PDIR_IDX(va);
787 	unsigned int ptbl_idx = PTBL_IDX(va);
788 	vm_page_t m;
789 	pte_t *ptbl;
790 	pte_t *pte;
791 
792 	//int su = (pmap == kernel_pmap);
793 	//debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
794 	//		su, (u_int32_t)pmap, va, flags);
795 
796 	ptbl = pmap->pm_pdir[pdir_idx];
797 	KASSERT(ptbl, ("pte_remove: null ptbl"));
798 
799 	pte = &ptbl[ptbl_idx];
800 
801 	if (pte == NULL || !PTE_ISVALID(pte))
802 		return (0);
803 
804 	if (PTE_ISWIRED(pte))
805 		pmap->pm_stats.wired_count--;
806 
807 	/* Handle managed entry. */
808 	if (PTE_ISMANAGED(pte)) {
809 		/* Get vm_page_t for mapped pte. */
810 		m = PHYS_TO_VM_PAGE(PTE_PA(pte));
811 
812 		if (PTE_ISMODIFIED(pte))
813 			vm_page_dirty(m);
814 
815 		if (PTE_ISREFERENCED(pte))
816 			vm_page_flag_set(m, PG_REFERENCED);
817 
818 		pv_remove(pmap, va, m);
819 	}
820 
821 	mtx_lock_spin(&tlbivax_mutex);
822 	tlb_miss_lock();
823 
824 	tlb0_flush_entry(va);
825 	pte->flags = 0;
826 	pte->rpn = 0;
827 
828 	tlb_miss_unlock();
829 	mtx_unlock_spin(&tlbivax_mutex);
830 
831 	pmap->pm_stats.resident_count--;
832 
833 	if (flags & PTBL_UNHOLD) {
834 		//debugf("pte_remove: e (unhold)\n");
835 		return (ptbl_unhold(mmu, pmap, pdir_idx));
836 	}
837 
838 	//debugf("pte_remove: e\n");
839 	return (0);
840 }
841 
842 /*
843  * Insert PTE for a given page and virtual address.
844  */
845 static void
846 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags)
847 {
848 	unsigned int pdir_idx = PDIR_IDX(va);
849 	unsigned int ptbl_idx = PTBL_IDX(va);
850 	pte_t *ptbl, *pte;
851 
852 	CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
853 	    pmap == kernel_pmap, pmap, va);
854 
855 	/* Get the page table pointer. */
856 	ptbl = pmap->pm_pdir[pdir_idx];
857 
858 	if (ptbl == NULL) {
859 		/* Allocate page table pages. */
860 		ptbl = ptbl_alloc(mmu, pmap, pdir_idx);
861 	} else {
862 		/*
863 		 * Check if there is valid mapping for requested
864 		 * va, if there is, remove it.
865 		 */
866 		pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
867 		if (PTE_ISVALID(pte)) {
868 			pte_remove(mmu, pmap, va, PTBL_HOLD);
869 		} else {
870 			/*
871 			 * pte is not used, increment hold count
872 			 * for ptbl pages.
873 			 */
874 			if (pmap != kernel_pmap)
875 				ptbl_hold(mmu, pmap, pdir_idx);
876 		}
877 	}
878 
879 	/*
880 	 * Insert pv_entry into pv_list for mapped page if part of managed
881 	 * memory.
882 	 */
883         if ((m->flags & PG_FICTITIOUS) == 0) {
884 		if ((m->flags & PG_UNMANAGED) == 0) {
885 			flags |= PTE_MANAGED;
886 
887 			/* Create and insert pv entry. */
888 			pv_insert(pmap, va, m);
889 		}
890 	}
891 
892 	pmap->pm_stats.resident_count++;
893 
894 	mtx_lock_spin(&tlbivax_mutex);
895 	tlb_miss_lock();
896 
897 	tlb0_flush_entry(va);
898 	if (pmap->pm_pdir[pdir_idx] == NULL) {
899 		/*
900 		 * If we just allocated a new page table, hook it in
901 		 * the pdir.
902 		 */
903 		pmap->pm_pdir[pdir_idx] = ptbl;
904 	}
905 	pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
906 	pte->rpn = VM_PAGE_TO_PHYS(m) & ~PTE_PA_MASK;
907 	pte->flags |= (PTE_VALID | flags);
908 
909 	tlb_miss_unlock();
910 	mtx_unlock_spin(&tlbivax_mutex);
911 }
912 
913 /* Return the pa for the given pmap/va. */
914 static vm_paddr_t
915 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
916 {
917 	vm_paddr_t pa = 0;
918 	pte_t *pte;
919 
920 	pte = pte_find(mmu, pmap, va);
921 	if ((pte != NULL) && PTE_ISVALID(pte))
922 		pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
923 	return (pa);
924 }
925 
926 /* Get a pointer to a PTE in a page table. */
927 static pte_t *
928 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
929 {
930 	unsigned int pdir_idx = PDIR_IDX(va);
931 	unsigned int ptbl_idx = PTBL_IDX(va);
932 
933 	KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
934 
935 	if (pmap->pm_pdir[pdir_idx])
936 		return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
937 
938 	return (NULL);
939 }
940 
941 /**************************************************************************/
942 /* PMAP related */
943 /**************************************************************************/
944 
945 /*
946  * This is called during booke_init, before the system is really initialized.
947  */
948 static void
949 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
950 {
951 	vm_offset_t phys_kernelend;
952 	struct mem_region *mp, *mp1;
953 	int cnt, i, j;
954 	u_int s, e, sz;
955 	u_int phys_avail_count;
956 	vm_size_t physsz, hwphyssz, kstack0_sz;
957 	vm_offset_t kernel_pdir, kstack0, va;
958 	vm_paddr_t kstack0_phys;
959 	void *dpcpu;
960 	pte_t *pte;
961 
962 	debugf("mmu_booke_bootstrap: entered\n");
963 
964 	/* Initialize invalidation mutex */
965 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
966 
967 	/* Read TLB0 size and associativity. */
968 	tlb0_get_tlbconf();
969 
970 	/* Align kernel start and end address (kernel image). */
971 	kernstart = trunc_page(start);
972 	data_start = round_page(kernelend);
973 	kernsize = data_start - kernstart;
974 
975 	data_end = data_start;
976 
977 	/* Allocate space for the message buffer. */
978 	msgbufp = (struct msgbuf *)data_end;
979 	data_end += msgbufsize;
980 	debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
981 	    data_end);
982 
983 	data_end = round_page(data_end);
984 
985 	/* Allocate the dynamic per-cpu area. */
986 	dpcpu = (void *)data_end;
987 	data_end += DPCPU_SIZE;
988 	dpcpu_init(dpcpu, 0);
989 
990 	/* Allocate space for ptbl_bufs. */
991 	ptbl_bufs = (struct ptbl_buf *)data_end;
992 	data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
993 	debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
994 	    data_end);
995 
996 	data_end = round_page(data_end);
997 
998 	/* Allocate PTE tables for kernel KVA. */
999 	kernel_pdir = data_end;
1000 	kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1001 	    PDIR_SIZE - 1) / PDIR_SIZE;
1002 	data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1003 	debugf(" kernel ptbls: %d\n", kernel_ptbls);
1004 	debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1005 
1006 	debugf(" data_end: 0x%08x\n", data_end);
1007 	if (data_end - kernstart > 0x1000000) {
1008 		data_end = (data_end + 0x3fffff) & ~0x3fffff;
1009 		tlb1_mapin_region(kernstart + 0x1000000,
1010 		    kernload + 0x1000000, data_end - kernstart - 0x1000000);
1011 	} else
1012 		data_end = (data_end + 0xffffff) & ~0xffffff;
1013 
1014 	debugf(" updated data_end: 0x%08x\n", data_end);
1015 
1016 	kernsize += data_end - data_start;
1017 
1018 	/*
1019 	 * Clear the structures - note we can only do it safely after the
1020 	 * possible additional TLB1 translations are in place (above) so that
1021 	 * all range up to the currently calculated 'data_end' is covered.
1022 	 */
1023 	memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1024 	memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1025 
1026 	/*******************************************************/
1027 	/* Set the start and end of kva. */
1028 	/*******************************************************/
1029 	virtual_avail = round_page(data_end);
1030 	virtual_end = VM_MAX_KERNEL_ADDRESS;
1031 
1032 	/* Allocate KVA space for page zero/copy operations. */
1033 	zero_page_va = virtual_avail;
1034 	virtual_avail += PAGE_SIZE;
1035 	zero_page_idle_va = virtual_avail;
1036 	virtual_avail += PAGE_SIZE;
1037 	copy_page_src_va = virtual_avail;
1038 	virtual_avail += PAGE_SIZE;
1039 	copy_page_dst_va = virtual_avail;
1040 	virtual_avail += PAGE_SIZE;
1041 	debugf("zero_page_va = 0x%08x\n", zero_page_va);
1042 	debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1043 	debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1044 	debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1045 
1046 	/* Initialize page zero/copy mutexes. */
1047 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1048 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1049 
1050 	/* Allocate KVA space for ptbl bufs. */
1051 	ptbl_buf_pool_vabase = virtual_avail;
1052 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1053 	debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1054 	    ptbl_buf_pool_vabase, virtual_avail);
1055 
1056 	/* Calculate corresponding physical addresses for the kernel region. */
1057 	phys_kernelend = kernload + kernsize;
1058 	debugf("kernel image and allocated data:\n");
1059 	debugf(" kernload    = 0x%08x\n", kernload);
1060 	debugf(" kernstart   = 0x%08x\n", kernstart);
1061 	debugf(" kernsize    = 0x%08x\n", kernsize);
1062 
1063 	if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1064 		panic("mmu_booke_bootstrap: phys_avail too small");
1065 
1066 	/*
1067 	 * Remove kernel physical address range from avail regions list. Page
1068 	 * align all regions.  Non-page aligned memory isn't very interesting
1069 	 * to us.  Also, sort the entries for ascending addresses.
1070 	 */
1071 
1072 	/* Retrieve phys/avail mem regions */
1073 	mem_regions(&physmem_regions, &physmem_regions_sz,
1074 	    &availmem_regions, &availmem_regions_sz);
1075 	sz = 0;
1076 	cnt = availmem_regions_sz;
1077 	debugf("processing avail regions:\n");
1078 	for (mp = availmem_regions; mp->mr_size; mp++) {
1079 		s = mp->mr_start;
1080 		e = mp->mr_start + mp->mr_size;
1081 		debugf(" %08x-%08x -> ", s, e);
1082 		/* Check whether this region holds all of the kernel. */
1083 		if (s < kernload && e > phys_kernelend) {
1084 			availmem_regions[cnt].mr_start = phys_kernelend;
1085 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
1086 			e = kernload;
1087 		}
1088 		/* Look whether this regions starts within the kernel. */
1089 		if (s >= kernload && s < phys_kernelend) {
1090 			if (e <= phys_kernelend)
1091 				goto empty;
1092 			s = phys_kernelend;
1093 		}
1094 		/* Now look whether this region ends within the kernel. */
1095 		if (e > kernload && e <= phys_kernelend) {
1096 			if (s >= kernload)
1097 				goto empty;
1098 			e = kernload;
1099 		}
1100 		/* Now page align the start and size of the region. */
1101 		s = round_page(s);
1102 		e = trunc_page(e);
1103 		if (e < s)
1104 			e = s;
1105 		sz = e - s;
1106 		debugf("%08x-%08x = %x\n", s, e, sz);
1107 
1108 		/* Check whether some memory is left here. */
1109 		if (sz == 0) {
1110 		empty:
1111 			memmove(mp, mp + 1,
1112 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
1113 			cnt--;
1114 			mp--;
1115 			continue;
1116 		}
1117 
1118 		/* Do an insertion sort. */
1119 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
1120 			if (s < mp1->mr_start)
1121 				break;
1122 		if (mp1 < mp) {
1123 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1124 			mp1->mr_start = s;
1125 			mp1->mr_size = sz;
1126 		} else {
1127 			mp->mr_start = s;
1128 			mp->mr_size = sz;
1129 		}
1130 	}
1131 	availmem_regions_sz = cnt;
1132 
1133 	/*******************************************************/
1134 	/* Steal physical memory for kernel stack from the end */
1135 	/* of the first avail region                           */
1136 	/*******************************************************/
1137 	kstack0_sz = KSTACK_PAGES * PAGE_SIZE;
1138 	kstack0_phys = availmem_regions[0].mr_start +
1139 	    availmem_regions[0].mr_size;
1140 	kstack0_phys -= kstack0_sz;
1141 	availmem_regions[0].mr_size -= kstack0_sz;
1142 
1143 	/*******************************************************/
1144 	/* Fill in phys_avail table, based on availmem_regions */
1145 	/*******************************************************/
1146 	phys_avail_count = 0;
1147 	physsz = 0;
1148 	hwphyssz = 0;
1149 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1150 
1151 	debugf("fill in phys_avail:\n");
1152 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1153 
1154 		debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1155 		    availmem_regions[i].mr_start,
1156 		    availmem_regions[i].mr_start +
1157 		        availmem_regions[i].mr_size,
1158 		    availmem_regions[i].mr_size);
1159 
1160 		if (hwphyssz != 0 &&
1161 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1162 			debugf(" hw.physmem adjust\n");
1163 			if (physsz < hwphyssz) {
1164 				phys_avail[j] = availmem_regions[i].mr_start;
1165 				phys_avail[j + 1] =
1166 				    availmem_regions[i].mr_start +
1167 				    hwphyssz - physsz;
1168 				physsz = hwphyssz;
1169 				phys_avail_count++;
1170 			}
1171 			break;
1172 		}
1173 
1174 		phys_avail[j] = availmem_regions[i].mr_start;
1175 		phys_avail[j + 1] = availmem_regions[i].mr_start +
1176 		    availmem_regions[i].mr_size;
1177 		phys_avail_count++;
1178 		physsz += availmem_regions[i].mr_size;
1179 	}
1180 	physmem = btoc(physsz);
1181 
1182 	/* Calculate the last available physical address. */
1183 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
1184 		;
1185 	Maxmem = powerpc_btop(phys_avail[i + 1]);
1186 
1187 	debugf("Maxmem = 0x%08lx\n", Maxmem);
1188 	debugf("phys_avail_count = %d\n", phys_avail_count);
1189 	debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem,
1190 	    physmem);
1191 
1192 	/*******************************************************/
1193 	/* Initialize (statically allocated) kernel pmap. */
1194 	/*******************************************************/
1195 	PMAP_LOCK_INIT(kernel_pmap);
1196 	kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1197 
1198 	debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1199 	debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1200 	debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1201 	    kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1202 
1203 	/* Initialize kernel pdir */
1204 	for (i = 0; i < kernel_ptbls; i++)
1205 		kernel_pmap->pm_pdir[kptbl_min + i] =
1206 		    (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1207 
1208 	for (i = 0; i < MAXCPU; i++) {
1209 		kernel_pmap->pm_tid[i] = TID_KERNEL;
1210 
1211 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1212 		tidbusy[i][0] = kernel_pmap;
1213 	}
1214 
1215 	/*
1216 	 * Fill in PTEs covering kernel code and data. They are not required
1217 	 * for address translation, as this area is covered by static TLB1
1218 	 * entries, but for pte_vatopa() to work correctly with kernel area
1219 	 * addresses.
1220 	 */
1221 	for (va = KERNBASE; va < data_end; va += PAGE_SIZE) {
1222 		pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1223 		pte->rpn = kernload + (va - KERNBASE);
1224 		pte->flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1225 		    PTE_VALID;
1226 	}
1227 	/* Mark kernel_pmap active on all CPUs */
1228 	kernel_pmap->pm_active = ~0;
1229 
1230 	/*******************************************************/
1231 	/* Final setup */
1232 	/*******************************************************/
1233 
1234 	/* Enter kstack0 into kernel map, provide guard page */
1235 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1236 	thread0.td_kstack = kstack0;
1237 	thread0.td_kstack_pages = KSTACK_PAGES;
1238 
1239 	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1240 	debugf("kstack0_phys at 0x%08x - 0x%08x\n",
1241 	    kstack0_phys, kstack0_phys + kstack0_sz);
1242 	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1243 
1244 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1245 	for (i = 0; i < KSTACK_PAGES; i++) {
1246 		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1247 		kstack0 += PAGE_SIZE;
1248 		kstack0_phys += PAGE_SIZE;
1249 	}
1250 
1251 	debugf("virtual_avail = %08x\n", virtual_avail);
1252 	debugf("virtual_end   = %08x\n", virtual_end);
1253 
1254 	debugf("mmu_booke_bootstrap: exit\n");
1255 }
1256 
1257 void
1258 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1259 {
1260 	int i;
1261 
1262 	/*
1263 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1264 	 * have the snapshot of its contents in the s/w tlb1[] table, so use
1265 	 * these values directly to (re)program AP's TLB1 hardware.
1266 	 */
1267 	for (i = 0; i < tlb1_idx; i ++) {
1268 		/* Skip invalid entries */
1269 		if (!(tlb1[i].mas1 & MAS1_VALID))
1270 			continue;
1271 
1272 		tlb1_write_entry(i);
1273 	}
1274 
1275 	set_mas4_defaults();
1276 }
1277 
1278 /*
1279  * Get the physical page address for the given pmap/virtual address.
1280  */
1281 static vm_paddr_t
1282 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1283 {
1284 	vm_paddr_t pa;
1285 
1286 	PMAP_LOCK(pmap);
1287 	pa = pte_vatopa(mmu, pmap, va);
1288 	PMAP_UNLOCK(pmap);
1289 
1290 	return (pa);
1291 }
1292 
1293 /*
1294  * Extract the physical page address associated with the given
1295  * kernel virtual address.
1296  */
1297 static vm_paddr_t
1298 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1299 {
1300 
1301 	return (pte_vatopa(mmu, kernel_pmap, va));
1302 }
1303 
1304 /*
1305  * Initialize the pmap module.
1306  * Called by vm_init, to initialize any structures that the pmap
1307  * system needs to map virtual memory.
1308  */
1309 static void
1310 mmu_booke_init(mmu_t mmu)
1311 {
1312 	int shpgperproc = PMAP_SHPGPERPROC;
1313 
1314 	/*
1315 	 * Initialize the address space (zone) for the pv entries.  Set a
1316 	 * high water mark so that the system can recover from excessive
1317 	 * numbers of pv entries.
1318 	 */
1319 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1320 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1321 
1322 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1323 	pv_entry_max = shpgperproc * maxproc + cnt.v_page_count;
1324 
1325 	TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1326 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1327 
1328 	uma_zone_set_obj(pvzone, &pvzone_obj, pv_entry_max);
1329 
1330 	/* Pre-fill pvzone with initial number of pv entries. */
1331 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1332 
1333 	/* Initialize ptbl allocation. */
1334 	ptbl_init();
1335 }
1336 
1337 /*
1338  * Map a list of wired pages into kernel virtual address space.  This is
1339  * intended for temporary mappings which do not need page modification or
1340  * references recorded.  Existing mappings in the region are overwritten.
1341  */
1342 static void
1343 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1344 {
1345 	vm_offset_t va;
1346 
1347 	va = sva;
1348 	while (count-- > 0) {
1349 		mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1350 		va += PAGE_SIZE;
1351 		m++;
1352 	}
1353 }
1354 
1355 /*
1356  * Remove page mappings from kernel virtual address space.  Intended for
1357  * temporary mappings entered by mmu_booke_qenter.
1358  */
1359 static void
1360 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1361 {
1362 	vm_offset_t va;
1363 
1364 	va = sva;
1365 	while (count-- > 0) {
1366 		mmu_booke_kremove(mmu, va);
1367 		va += PAGE_SIZE;
1368 	}
1369 }
1370 
1371 /*
1372  * Map a wired page into kernel virtual address space.
1373  */
1374 static void
1375 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_offset_t pa)
1376 {
1377 	unsigned int pdir_idx = PDIR_IDX(va);
1378 	unsigned int ptbl_idx = PTBL_IDX(va);
1379 	uint32_t flags;
1380 	pte_t *pte;
1381 
1382 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1383 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1384 
1385 	flags = 0;
1386 	flags |= (PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID);
1387 	flags |= PTE_M;
1388 
1389 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1390 
1391 	mtx_lock_spin(&tlbivax_mutex);
1392 	tlb_miss_lock();
1393 
1394 	if (PTE_ISVALID(pte)) {
1395 
1396 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1397 
1398 		/* Flush entry from TLB0 */
1399 		tlb0_flush_entry(va);
1400 	}
1401 
1402 	pte->rpn = pa & ~PTE_PA_MASK;
1403 	pte->flags = flags;
1404 
1405 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1406 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1407 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1408 
1409 	/* Flush the real memory from the instruction cache. */
1410 	if ((flags & (PTE_I | PTE_G)) == 0) {
1411 		__syncicache((void *)va, PAGE_SIZE);
1412 	}
1413 
1414 	tlb_miss_unlock();
1415 	mtx_unlock_spin(&tlbivax_mutex);
1416 }
1417 
1418 /*
1419  * Remove a page from kernel page table.
1420  */
1421 static void
1422 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1423 {
1424 	unsigned int pdir_idx = PDIR_IDX(va);
1425 	unsigned int ptbl_idx = PTBL_IDX(va);
1426 	pte_t *pte;
1427 
1428 //	CTR2(KTR_PMAP,("%s: s (va = 0x%08x)\n", __func__, va));
1429 
1430 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1431 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1432 	    ("mmu_booke_kremove: invalid va"));
1433 
1434 	pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1435 
1436 	if (!PTE_ISVALID(pte)) {
1437 
1438 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1439 
1440 		return;
1441 	}
1442 
1443 	mtx_lock_spin(&tlbivax_mutex);
1444 	tlb_miss_lock();
1445 
1446 	/* Invalidate entry in TLB0, update PTE. */
1447 	tlb0_flush_entry(va);
1448 	pte->flags = 0;
1449 	pte->rpn = 0;
1450 
1451 	tlb_miss_unlock();
1452 	mtx_unlock_spin(&tlbivax_mutex);
1453 }
1454 
1455 /*
1456  * Initialize pmap associated with process 0.
1457  */
1458 static void
1459 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1460 {
1461 
1462 	mmu_booke_pinit(mmu, pmap);
1463 	PCPU_SET(curpmap, pmap);
1464 }
1465 
1466 /*
1467  * Initialize a preallocated and zeroed pmap structure,
1468  * such as one in a vmspace structure.
1469  */
1470 static void
1471 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1472 {
1473 	int i;
1474 
1475 	CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1476 	    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1477 
1478 	KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1479 
1480 	PMAP_LOCK_INIT(pmap);
1481 	for (i = 0; i < MAXCPU; i++)
1482 		pmap->pm_tid[i] = TID_NONE;
1483 	pmap->pm_active = 0;
1484 	bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1485 	bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1486 	TAILQ_INIT(&pmap->pm_ptbl_list);
1487 }
1488 
1489 /*
1490  * Release any resources held by the given physical map.
1491  * Called when a pmap initialized by mmu_booke_pinit is being released.
1492  * Should only be called if the map contains no valid mappings.
1493  */
1494 static void
1495 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1496 {
1497 
1498 	KASSERT(pmap->pm_stats.resident_count == 0,
1499 	    ("pmap_release: pmap resident count %ld != 0",
1500 	    pmap->pm_stats.resident_count));
1501 
1502 	PMAP_LOCK_DESTROY(pmap);
1503 }
1504 
1505 /*
1506  * Insert the given physical page at the specified virtual address in the
1507  * target physical map with the protection requested. If specified the page
1508  * will be wired down.
1509  */
1510 static void
1511 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1512     vm_prot_t prot, boolean_t wired)
1513 {
1514 
1515 	vm_page_lock_queues();
1516 	PMAP_LOCK(pmap);
1517 	mmu_booke_enter_locked(mmu, pmap, va, m, prot, wired);
1518 	vm_page_unlock_queues();
1519 	PMAP_UNLOCK(pmap);
1520 }
1521 
1522 static void
1523 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1524     vm_prot_t prot, boolean_t wired)
1525 {
1526 	pte_t *pte;
1527 	vm_paddr_t pa;
1528 	uint32_t flags;
1529 	int su, sync;
1530 
1531 	pa = VM_PAGE_TO_PHYS(m);
1532 	su = (pmap == kernel_pmap);
1533 	sync = 0;
1534 
1535 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1536 	//		"pa=0x%08x prot=0x%08x wired=%d)\n",
1537 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1538 	//		(u_int32_t)m, va, pa, prot, wired);
1539 
1540 	if (su) {
1541 		KASSERT(((va >= virtual_avail) &&
1542 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1543 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1544 	} else {
1545 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1546 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1547 	}
1548 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0 ||
1549 	    (m->oflags & VPO_BUSY) != 0 || VM_OBJECT_LOCKED(m->object),
1550 	    ("mmu_booke_enter_locked: page %p is not busy", m));
1551 
1552 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1553 
1554 	/*
1555 	 * If there is an existing mapping, and the physical address has not
1556 	 * changed, must be protection or wiring change.
1557 	 */
1558 	if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1559 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1560 
1561 		/*
1562 		 * Before actually updating pte->flags we calculate and
1563 		 * prepare its new value in a helper var.
1564 		 */
1565 		flags = pte->flags;
1566 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1567 
1568 		/* Wiring change, just update stats. */
1569 		if (wired) {
1570 			if (!PTE_ISWIRED(pte)) {
1571 				flags |= PTE_WIRED;
1572 				pmap->pm_stats.wired_count++;
1573 			}
1574 		} else {
1575 			if (PTE_ISWIRED(pte)) {
1576 				flags &= ~PTE_WIRED;
1577 				pmap->pm_stats.wired_count--;
1578 			}
1579 		}
1580 
1581 		if (prot & VM_PROT_WRITE) {
1582 			/* Add write permissions. */
1583 			flags |= PTE_SW;
1584 			if (!su)
1585 				flags |= PTE_UW;
1586 
1587 			if ((flags & PTE_MANAGED) != 0)
1588 				vm_page_flag_set(m, PG_WRITEABLE);
1589 		} else {
1590 			/* Handle modified pages, sense modify status. */
1591 
1592 			/*
1593 			 * The PTE_MODIFIED flag could be set by underlying
1594 			 * TLB misses since we last read it (above), possibly
1595 			 * other CPUs could update it so we check in the PTE
1596 			 * directly rather than rely on that saved local flags
1597 			 * copy.
1598 			 */
1599 			if (PTE_ISMODIFIED(pte))
1600 				vm_page_dirty(m);
1601 		}
1602 
1603 		if (prot & VM_PROT_EXECUTE) {
1604 			flags |= PTE_SX;
1605 			if (!su)
1606 				flags |= PTE_UX;
1607 
1608 			/*
1609 			 * Check existing flags for execute permissions: if we
1610 			 * are turning execute permissions on, icache should
1611 			 * be flushed.
1612 			 */
1613 			if ((pte->flags & (PTE_UX | PTE_SX)) == 0)
1614 				sync++;
1615 		}
1616 
1617 		flags &= ~PTE_REFERENCED;
1618 
1619 		/*
1620 		 * The new flags value is all calculated -- only now actually
1621 		 * update the PTE.
1622 		 */
1623 		mtx_lock_spin(&tlbivax_mutex);
1624 		tlb_miss_lock();
1625 
1626 		tlb0_flush_entry(va);
1627 		pte->flags = flags;
1628 
1629 		tlb_miss_unlock();
1630 		mtx_unlock_spin(&tlbivax_mutex);
1631 
1632 	} else {
1633 		/*
1634 		 * If there is an existing mapping, but it's for a different
1635 		 * physical address, pte_enter() will delete the old mapping.
1636 		 */
1637 		//if ((pte != NULL) && PTE_ISVALID(pte))
1638 		//	debugf("mmu_booke_enter_locked: replace\n");
1639 		//else
1640 		//	debugf("mmu_booke_enter_locked: new\n");
1641 
1642 		/* Now set up the flags and install the new mapping. */
1643 		flags = (PTE_SR | PTE_VALID);
1644 		flags |= PTE_M;
1645 
1646 		if (!su)
1647 			flags |= PTE_UR;
1648 
1649 		if (prot & VM_PROT_WRITE) {
1650 			flags |= PTE_SW;
1651 			if (!su)
1652 				flags |= PTE_UW;
1653 
1654 			if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0)
1655 				vm_page_flag_set(m, PG_WRITEABLE);
1656 		}
1657 
1658 		if (prot & VM_PROT_EXECUTE) {
1659 			flags |= PTE_SX;
1660 			if (!su)
1661 				flags |= PTE_UX;
1662 		}
1663 
1664 		/* If its wired update stats. */
1665 		if (wired) {
1666 			pmap->pm_stats.wired_count++;
1667 			flags |= PTE_WIRED;
1668 		}
1669 
1670 		pte_enter(mmu, pmap, m, va, flags);
1671 
1672 		/* Flush the real memory from the instruction cache. */
1673 		if (prot & VM_PROT_EXECUTE)
1674 			sync++;
1675 	}
1676 
1677 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1678 		__syncicache((void *)va, PAGE_SIZE);
1679 		sync = 0;
1680 	}
1681 }
1682 
1683 /*
1684  * Maps a sequence of resident pages belonging to the same object.
1685  * The sequence begins with the given page m_start.  This page is
1686  * mapped at the given virtual address start.  Each subsequent page is
1687  * mapped at a virtual address that is offset from start by the same
1688  * amount as the page is offset from m_start within the object.  The
1689  * last page in the sequence is the page with the largest offset from
1690  * m_start that can be mapped at a virtual address less than the given
1691  * virtual address end.  Not every virtual page between start and end
1692  * is mapped; only those for which a resident page exists with the
1693  * corresponding offset from m_start are mapped.
1694  */
1695 static void
1696 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1697     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1698 {
1699 	vm_page_t m;
1700 	vm_pindex_t diff, psize;
1701 
1702 	psize = atop(end - start);
1703 	m = m_start;
1704 	vm_page_lock_queues();
1705 	PMAP_LOCK(pmap);
1706 	while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1707 		mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1708 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1709 		m = TAILQ_NEXT(m, listq);
1710 	}
1711 	vm_page_unlock_queues();
1712 	PMAP_UNLOCK(pmap);
1713 }
1714 
1715 static void
1716 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1717     vm_prot_t prot)
1718 {
1719 
1720 	vm_page_lock_queues();
1721 	PMAP_LOCK(pmap);
1722 	mmu_booke_enter_locked(mmu, pmap, va, m,
1723 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE);
1724 	vm_page_unlock_queues();
1725 	PMAP_UNLOCK(pmap);
1726 }
1727 
1728 /*
1729  * Remove the given range of addresses from the specified map.
1730  *
1731  * It is assumed that the start and end are properly rounded to the page size.
1732  */
1733 static void
1734 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1735 {
1736 	pte_t *pte;
1737 	uint8_t hold_flag;
1738 
1739 	int su = (pmap == kernel_pmap);
1740 
1741 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1742 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1743 
1744 	if (su) {
1745 		KASSERT(((va >= virtual_avail) &&
1746 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1747 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1748 	} else {
1749 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1750 		    ("mmu_booke_remove: user pmap, non user va"));
1751 	}
1752 
1753 	if (PMAP_REMOVE_DONE(pmap)) {
1754 		//debugf("mmu_booke_remove: e (empty)\n");
1755 		return;
1756 	}
1757 
1758 	hold_flag = PTBL_HOLD_FLAG(pmap);
1759 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1760 
1761 	vm_page_lock_queues();
1762 	PMAP_LOCK(pmap);
1763 	for (; va < endva; va += PAGE_SIZE) {
1764 		pte = pte_find(mmu, pmap, va);
1765 		if ((pte != NULL) && PTE_ISVALID(pte))
1766 			pte_remove(mmu, pmap, va, hold_flag);
1767 	}
1768 	PMAP_UNLOCK(pmap);
1769 	vm_page_unlock_queues();
1770 
1771 	//debugf("mmu_booke_remove: e\n");
1772 }
1773 
1774 /*
1775  * Remove physical page from all pmaps in which it resides.
1776  */
1777 static void
1778 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1779 {
1780 	pv_entry_t pv, pvn;
1781 	uint8_t hold_flag;
1782 
1783 	vm_page_lock_queues();
1784 	for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1785 		pvn = TAILQ_NEXT(pv, pv_link);
1786 
1787 		PMAP_LOCK(pv->pv_pmap);
1788 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1789 		pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1790 		PMAP_UNLOCK(pv->pv_pmap);
1791 	}
1792 	vm_page_flag_clear(m, PG_WRITEABLE);
1793 	vm_page_unlock_queues();
1794 }
1795 
1796 /*
1797  * Map a range of physical addresses into kernel virtual address space.
1798  */
1799 static vm_offset_t
1800 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_offset_t pa_start,
1801     vm_offset_t pa_end, int prot)
1802 {
1803 	vm_offset_t sva = *virt;
1804 	vm_offset_t va = sva;
1805 
1806 	//debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1807 	//		sva, pa_start, pa_end);
1808 
1809 	while (pa_start < pa_end) {
1810 		mmu_booke_kenter(mmu, va, pa_start);
1811 		va += PAGE_SIZE;
1812 		pa_start += PAGE_SIZE;
1813 	}
1814 	*virt = va;
1815 
1816 	//debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1817 	return (sva);
1818 }
1819 
1820 /*
1821  * The pmap must be activated before it's address space can be accessed in any
1822  * way.
1823  */
1824 static void
1825 mmu_booke_activate(mmu_t mmu, struct thread *td)
1826 {
1827 	pmap_t pmap;
1828 
1829 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1830 
1831 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1832 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1833 
1834 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1835 
1836 	mtx_lock_spin(&sched_lock);
1837 
1838 	atomic_set_int(&pmap->pm_active, PCPU_GET(cpumask));
1839 	PCPU_SET(curpmap, pmap);
1840 
1841 	if (pmap->pm_tid[PCPU_GET(cpuid)] == TID_NONE)
1842 		tid_alloc(pmap);
1843 
1844 	/* Load PID0 register with pmap tid value. */
1845 	mtspr(SPR_PID0, pmap->pm_tid[PCPU_GET(cpuid)]);
1846 	__asm __volatile("isync");
1847 
1848 	mtx_unlock_spin(&sched_lock);
1849 
1850 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1851 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1852 }
1853 
1854 /*
1855  * Deactivate the specified process's address space.
1856  */
1857 static void
1858 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1859 {
1860 	pmap_t pmap;
1861 
1862 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1863 
1864 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
1865 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1866 
1867 	atomic_clear_int(&pmap->pm_active, PCPU_GET(cpumask));
1868 	PCPU_SET(curpmap, NULL);
1869 }
1870 
1871 /*
1872  * Copy the range specified by src_addr/len
1873  * from the source map to the range dst_addr/len
1874  * in the destination map.
1875  *
1876  * This routine is only advisory and need not do anything.
1877  */
1878 static void
1879 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
1880     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
1881 {
1882 
1883 }
1884 
1885 /*
1886  * Set the physical protection on the specified range of this map as requested.
1887  */
1888 static void
1889 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1890     vm_prot_t prot)
1891 {
1892 	vm_offset_t va;
1893 	vm_page_t m;
1894 	pte_t *pte;
1895 
1896 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1897 		mmu_booke_remove(mmu, pmap, sva, eva);
1898 		return;
1899 	}
1900 
1901 	if (prot & VM_PROT_WRITE)
1902 		return;
1903 
1904 	vm_page_lock_queues();
1905 	PMAP_LOCK(pmap);
1906 	for (va = sva; va < eva; va += PAGE_SIZE) {
1907 		if ((pte = pte_find(mmu, pmap, va)) != NULL) {
1908 			if (PTE_ISVALID(pte)) {
1909 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1910 
1911 				mtx_lock_spin(&tlbivax_mutex);
1912 				tlb_miss_lock();
1913 
1914 				/* Handle modified pages. */
1915 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
1916 					vm_page_dirty(m);
1917 
1918 				tlb0_flush_entry(va);
1919 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1920 
1921 				tlb_miss_unlock();
1922 				mtx_unlock_spin(&tlbivax_mutex);
1923 			}
1924 		}
1925 	}
1926 	PMAP_UNLOCK(pmap);
1927 	vm_page_unlock_queues();
1928 }
1929 
1930 /*
1931  * Clear the write and modified bits in each of the given page's mappings.
1932  */
1933 static void
1934 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
1935 {
1936 	pv_entry_t pv;
1937 	pte_t *pte;
1938 
1939 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
1940 	    ("mmu_booke_remove_write: page %p is not managed", m));
1941 
1942 	/*
1943 	 * If the page is not VPO_BUSY, then PG_WRITEABLE cannot be set by
1944 	 * another thread while the object is locked.  Thus, if PG_WRITEABLE
1945 	 * is clear, no page table entries need updating.
1946 	 */
1947 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
1948 	if ((m->oflags & VPO_BUSY) == 0 &&
1949 	    (m->flags & PG_WRITEABLE) == 0)
1950 		return;
1951 	vm_page_lock_queues();
1952 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1953 		PMAP_LOCK(pv->pv_pmap);
1954 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
1955 			if (PTE_ISVALID(pte)) {
1956 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1957 
1958 				mtx_lock_spin(&tlbivax_mutex);
1959 				tlb_miss_lock();
1960 
1961 				/* Handle modified pages. */
1962 				if (PTE_ISMODIFIED(pte))
1963 					vm_page_dirty(m);
1964 
1965 				/* Flush mapping from TLB0. */
1966 				pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1967 
1968 				tlb_miss_unlock();
1969 				mtx_unlock_spin(&tlbivax_mutex);
1970 			}
1971 		}
1972 		PMAP_UNLOCK(pv->pv_pmap);
1973 	}
1974 	vm_page_flag_clear(m, PG_WRITEABLE);
1975 	vm_page_unlock_queues();
1976 }
1977 
1978 static void
1979 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
1980 {
1981 	pte_t *pte;
1982 	pmap_t pmap;
1983 	vm_page_t m;
1984 	vm_offset_t addr;
1985 	vm_paddr_t pa;
1986 	int active, valid;
1987 
1988 	va = trunc_page(va);
1989 	sz = round_page(sz);
1990 
1991 	vm_page_lock_queues();
1992 	pmap = PCPU_GET(curpmap);
1993 	active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
1994 	while (sz > 0) {
1995 		PMAP_LOCK(pm);
1996 		pte = pte_find(mmu, pm, va);
1997 		valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
1998 		if (valid)
1999 			pa = PTE_PA(pte);
2000 		PMAP_UNLOCK(pm);
2001 		if (valid) {
2002 			if (!active) {
2003 				/* Create a mapping in the active pmap. */
2004 				addr = 0;
2005 				m = PHYS_TO_VM_PAGE(pa);
2006 				PMAP_LOCK(pmap);
2007 				pte_enter(mmu, pmap, m, addr,
2008 				    PTE_SR | PTE_VALID | PTE_UR);
2009 				__syncicache((void *)addr, PAGE_SIZE);
2010 				pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2011 				PMAP_UNLOCK(pmap);
2012 			} else
2013 				__syncicache((void *)va, PAGE_SIZE);
2014 		}
2015 		va += PAGE_SIZE;
2016 		sz -= PAGE_SIZE;
2017 	}
2018 	vm_page_unlock_queues();
2019 }
2020 
2021 /*
2022  * Atomically extract and hold the physical page with the given
2023  * pmap and virtual address pair if that mapping permits the given
2024  * protection.
2025  */
2026 static vm_page_t
2027 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2028     vm_prot_t prot)
2029 {
2030 	pte_t *pte;
2031 	vm_page_t m;
2032 	uint32_t pte_wbit;
2033 	vm_paddr_t pa;
2034 
2035 	m = NULL;
2036 	pa = 0;
2037 	PMAP_LOCK(pmap);
2038 retry:
2039 	pte = pte_find(mmu, pmap, va);
2040 	if ((pte != NULL) && PTE_ISVALID(pte)) {
2041 		if (pmap == kernel_pmap)
2042 			pte_wbit = PTE_SW;
2043 		else
2044 			pte_wbit = PTE_UW;
2045 
2046 		if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2047 			if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2048 				goto retry;
2049 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2050 			vm_page_hold(m);
2051 		}
2052 	}
2053 
2054 	PA_UNLOCK_COND(pa);
2055 	PMAP_UNLOCK(pmap);
2056 	return (m);
2057 }
2058 
2059 /*
2060  * Initialize a vm_page's machine-dependent fields.
2061  */
2062 static void
2063 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2064 {
2065 
2066 	TAILQ_INIT(&m->md.pv_list);
2067 }
2068 
2069 /*
2070  * mmu_booke_zero_page_area zeros the specified hardware page by
2071  * mapping it into virtual memory and using bzero to clear
2072  * its contents.
2073  *
2074  * off and size must reside within a single page.
2075  */
2076 static void
2077 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2078 {
2079 	vm_offset_t va;
2080 
2081 	/* XXX KASSERT off and size are within a single page? */
2082 
2083 	mtx_lock(&zero_page_mutex);
2084 	va = zero_page_va;
2085 
2086 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2087 	bzero((caddr_t)va + off, size);
2088 	mmu_booke_kremove(mmu, va);
2089 
2090 	mtx_unlock(&zero_page_mutex);
2091 }
2092 
2093 /*
2094  * mmu_booke_zero_page zeros the specified hardware page.
2095  */
2096 static void
2097 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2098 {
2099 
2100 	mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
2101 }
2102 
2103 /*
2104  * mmu_booke_copy_page copies the specified (machine independent) page by
2105  * mapping the page into virtual memory and using memcopy to copy the page,
2106  * one machine dependent page at a time.
2107  */
2108 static void
2109 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2110 {
2111 	vm_offset_t sva, dva;
2112 
2113 	sva = copy_page_src_va;
2114 	dva = copy_page_dst_va;
2115 
2116 	mtx_lock(&copy_page_mutex);
2117 	mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2118 	mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2119 	memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2120 	mmu_booke_kremove(mmu, dva);
2121 	mmu_booke_kremove(mmu, sva);
2122 	mtx_unlock(&copy_page_mutex);
2123 }
2124 
2125 /*
2126  * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2127  * into virtual memory and using bzero to clear its contents. This is intended
2128  * to be called from the vm_pagezero process only and outside of Giant. No
2129  * lock is required.
2130  */
2131 static void
2132 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2133 {
2134 	vm_offset_t va;
2135 
2136 	va = zero_page_idle_va;
2137 	mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2138 	bzero((caddr_t)va, PAGE_SIZE);
2139 	mmu_booke_kremove(mmu, va);
2140 }
2141 
2142 /*
2143  * Return whether or not the specified physical page was modified
2144  * in any of physical maps.
2145  */
2146 static boolean_t
2147 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2148 {
2149 	pte_t *pte;
2150 	pv_entry_t pv;
2151 	boolean_t rv;
2152 
2153 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2154 	    ("mmu_booke_is_modified: page %p is not managed", m));
2155 	rv = FALSE;
2156 
2157 	/*
2158 	 * If the page is not VPO_BUSY, then PG_WRITEABLE cannot be
2159 	 * concurrently set while the object is locked.  Thus, if PG_WRITEABLE
2160 	 * is clear, no PTEs can be modified.
2161 	 */
2162 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2163 	if ((m->oflags & VPO_BUSY) == 0 &&
2164 	    (m->flags & PG_WRITEABLE) == 0)
2165 		return (rv);
2166 	vm_page_lock_queues();
2167 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2168 		PMAP_LOCK(pv->pv_pmap);
2169 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2170 		    PTE_ISVALID(pte)) {
2171 			if (PTE_ISMODIFIED(pte))
2172 				rv = TRUE;
2173 		}
2174 		PMAP_UNLOCK(pv->pv_pmap);
2175 		if (rv)
2176 			break;
2177 	}
2178 	vm_page_unlock_queues();
2179 	return (rv);
2180 }
2181 
2182 /*
2183  * Return whether or not the specified virtual address is eligible
2184  * for prefault.
2185  */
2186 static boolean_t
2187 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2188 {
2189 
2190 	return (FALSE);
2191 }
2192 
2193 /*
2194  * Return whether or not the specified physical page was referenced
2195  * in any physical maps.
2196  */
2197 static boolean_t
2198 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2199 {
2200 	pte_t *pte;
2201 	pv_entry_t pv;
2202 	boolean_t rv;
2203 
2204 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2205 	    ("mmu_booke_is_referenced: page %p is not managed", m));
2206 	rv = FALSE;
2207 	vm_page_lock_queues();
2208 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2209 		PMAP_LOCK(pv->pv_pmap);
2210 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2211 		    PTE_ISVALID(pte)) {
2212 			if (PTE_ISREFERENCED(pte))
2213 				rv = TRUE;
2214 		}
2215 		PMAP_UNLOCK(pv->pv_pmap);
2216 		if (rv)
2217 			break;
2218 	}
2219 	vm_page_unlock_queues();
2220 	return (rv);
2221 }
2222 
2223 /*
2224  * Clear the modify bits on the specified physical page.
2225  */
2226 static void
2227 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2228 {
2229 	pte_t *pte;
2230 	pv_entry_t pv;
2231 
2232 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2233 	    ("mmu_booke_clear_modify: page %p is not managed", m));
2234 	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
2235 	KASSERT((m->oflags & VPO_BUSY) == 0,
2236 	    ("mmu_booke_clear_modify: page %p is busy", m));
2237 
2238 	/*
2239 	 * If the page is not PG_WRITEABLE, then no PTEs can be modified.
2240 	 * If the object containing the page is locked and the page is not
2241 	 * VPO_BUSY, then PG_WRITEABLE cannot be concurrently set.
2242 	 */
2243 	if ((m->flags & PG_WRITEABLE) == 0)
2244 		return;
2245 	vm_page_lock_queues();
2246 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2247 		PMAP_LOCK(pv->pv_pmap);
2248 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2249 		    PTE_ISVALID(pte)) {
2250 			mtx_lock_spin(&tlbivax_mutex);
2251 			tlb_miss_lock();
2252 
2253 			if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2254 				tlb0_flush_entry(pv->pv_va);
2255 				pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2256 				    PTE_REFERENCED);
2257 			}
2258 
2259 			tlb_miss_unlock();
2260 			mtx_unlock_spin(&tlbivax_mutex);
2261 		}
2262 		PMAP_UNLOCK(pv->pv_pmap);
2263 	}
2264 	vm_page_unlock_queues();
2265 }
2266 
2267 /*
2268  * Return a count of reference bits for a page, clearing those bits.
2269  * It is not necessary for every reference bit to be cleared, but it
2270  * is necessary that 0 only be returned when there are truly no
2271  * reference bits set.
2272  *
2273  * XXX: The exact number of bits to check and clear is a matter that
2274  * should be tested and standardized at some point in the future for
2275  * optimal aging of shared pages.
2276  */
2277 static int
2278 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2279 {
2280 	pte_t *pte;
2281 	pv_entry_t pv;
2282 	int count;
2283 
2284 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2285 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
2286 	count = 0;
2287 	vm_page_lock_queues();
2288 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2289 		PMAP_LOCK(pv->pv_pmap);
2290 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2291 		    PTE_ISVALID(pte)) {
2292 			if (PTE_ISREFERENCED(pte)) {
2293 				mtx_lock_spin(&tlbivax_mutex);
2294 				tlb_miss_lock();
2295 
2296 				tlb0_flush_entry(pv->pv_va);
2297 				pte->flags &= ~PTE_REFERENCED;
2298 
2299 				tlb_miss_unlock();
2300 				mtx_unlock_spin(&tlbivax_mutex);
2301 
2302 				if (++count > 4) {
2303 					PMAP_UNLOCK(pv->pv_pmap);
2304 					break;
2305 				}
2306 			}
2307 		}
2308 		PMAP_UNLOCK(pv->pv_pmap);
2309 	}
2310 	vm_page_unlock_queues();
2311 	return (count);
2312 }
2313 
2314 /*
2315  * Clear the reference bit on the specified physical page.
2316  */
2317 static void
2318 mmu_booke_clear_reference(mmu_t mmu, vm_page_t m)
2319 {
2320 	pte_t *pte;
2321 	pv_entry_t pv;
2322 
2323 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2324 	    ("mmu_booke_clear_reference: page %p is not managed", m));
2325 	vm_page_lock_queues();
2326 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2327 		PMAP_LOCK(pv->pv_pmap);
2328 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2329 		    PTE_ISVALID(pte)) {
2330 			if (PTE_ISREFERENCED(pte)) {
2331 				mtx_lock_spin(&tlbivax_mutex);
2332 				tlb_miss_lock();
2333 
2334 				tlb0_flush_entry(pv->pv_va);
2335 				pte->flags &= ~PTE_REFERENCED;
2336 
2337 				tlb_miss_unlock();
2338 				mtx_unlock_spin(&tlbivax_mutex);
2339 			}
2340 		}
2341 		PMAP_UNLOCK(pv->pv_pmap);
2342 	}
2343 	vm_page_unlock_queues();
2344 }
2345 
2346 /*
2347  * Change wiring attribute for a map/virtual-address pair.
2348  */
2349 static void
2350 mmu_booke_change_wiring(mmu_t mmu, pmap_t pmap, vm_offset_t va, boolean_t wired)
2351 {
2352 	pte_t *pte;
2353 
2354 	PMAP_LOCK(pmap);
2355 	if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2356 		if (wired) {
2357 			if (!PTE_ISWIRED(pte)) {
2358 				pte->flags |= PTE_WIRED;
2359 				pmap->pm_stats.wired_count++;
2360 			}
2361 		} else {
2362 			if (PTE_ISWIRED(pte)) {
2363 				pte->flags &= ~PTE_WIRED;
2364 				pmap->pm_stats.wired_count--;
2365 			}
2366 		}
2367 	}
2368 	PMAP_UNLOCK(pmap);
2369 }
2370 
2371 /*
2372  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2373  * page.  This count may be changed upwards or downwards in the future; it is
2374  * only necessary that true be returned for a small subset of pmaps for proper
2375  * page aging.
2376  */
2377 static boolean_t
2378 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2379 {
2380 	pv_entry_t pv;
2381 	int loops;
2382 	boolean_t rv;
2383 
2384 	KASSERT((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0,
2385 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2386 	loops = 0;
2387 	rv = FALSE;
2388 	vm_page_lock_queues();
2389 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2390 		if (pv->pv_pmap == pmap) {
2391 			rv = TRUE;
2392 			break;
2393 		}
2394 		if (++loops >= 16)
2395 			break;
2396 	}
2397 	vm_page_unlock_queues();
2398 	return (rv);
2399 }
2400 
2401 /*
2402  * Return the number of managed mappings to the given physical page that are
2403  * wired.
2404  */
2405 static int
2406 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2407 {
2408 	pv_entry_t pv;
2409 	pte_t *pte;
2410 	int count = 0;
2411 
2412 	if ((m->flags & PG_FICTITIOUS) != 0)
2413 		return (count);
2414 	vm_page_lock_queues();
2415 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2416 		PMAP_LOCK(pv->pv_pmap);
2417 		if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2418 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2419 				count++;
2420 		PMAP_UNLOCK(pv->pv_pmap);
2421 	}
2422 	vm_page_unlock_queues();
2423 	return (count);
2424 }
2425 
2426 static int
2427 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2428 {
2429 	int i;
2430 	vm_offset_t va;
2431 
2432 	/*
2433 	 * This currently does not work for entries that
2434 	 * overlap TLB1 entries.
2435 	 */
2436 	for (i = 0; i < tlb1_idx; i ++) {
2437 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2438 			return (0);
2439 	}
2440 
2441 	return (EFAULT);
2442 }
2443 
2444 vm_offset_t
2445 mmu_booke_dumpsys_map(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2446     vm_size_t *sz)
2447 {
2448 	vm_paddr_t pa, ppa;
2449 	vm_offset_t va;
2450 	vm_size_t gran;
2451 
2452 	/* Raw physical memory dumps don't have a virtual address. */
2453 	if (md->md_vaddr == ~0UL) {
2454 		/* We always map a 256MB page at 256M. */
2455 		gran = 256 * 1024 * 1024;
2456 		pa = md->md_paddr + ofs;
2457 		ppa = pa & ~(gran - 1);
2458 		ofs = pa - ppa;
2459 		va = gran;
2460 		tlb1_set_entry(va, ppa, gran, _TLB_ENTRY_IO);
2461 		if (*sz > (gran - ofs))
2462 			*sz = gran - ofs;
2463 		return (va + ofs);
2464 	}
2465 
2466 	/* Minidumps are based on virtual memory addresses. */
2467 	va = md->md_vaddr + ofs;
2468 	if (va >= kernstart + kernsize) {
2469 		gran = PAGE_SIZE - (va & PAGE_MASK);
2470 		if (*sz > gran)
2471 			*sz = gran;
2472 	}
2473 	return (va);
2474 }
2475 
2476 void
2477 mmu_booke_dumpsys_unmap(mmu_t mmu, struct pmap_md *md, vm_size_t ofs,
2478     vm_offset_t va)
2479 {
2480 
2481 	/* Raw physical memory dumps don't have a virtual address. */
2482 	if (md->md_vaddr == ~0UL) {
2483 		tlb1_idx--;
2484 		tlb1[tlb1_idx].mas1 = 0;
2485 		tlb1[tlb1_idx].mas2 = 0;
2486 		tlb1[tlb1_idx].mas3 = 0;
2487 		tlb1_write_entry(tlb1_idx);
2488 		return;
2489 	}
2490 
2491 	/* Minidumps are based on virtual memory addresses. */
2492 	/* Nothing to do... */
2493 }
2494 
2495 struct pmap_md *
2496 mmu_booke_scan_md(mmu_t mmu, struct pmap_md *prev)
2497 {
2498 	static struct pmap_md md;
2499 	pte_t *pte;
2500 	vm_offset_t va;
2501 
2502 	if (dumpsys_minidump) {
2503 		md.md_paddr = ~0UL;	/* Minidumps use virtual addresses. */
2504 		if (prev == NULL) {
2505 			/* 1st: kernel .data and .bss. */
2506 			md.md_index = 1;
2507 			md.md_vaddr = trunc_page((uintptr_t)_etext);
2508 			md.md_size = round_page((uintptr_t)_end) - md.md_vaddr;
2509 			return (&md);
2510 		}
2511 		switch (prev->md_index) {
2512 		case 1:
2513 			/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2514 			md.md_index = 2;
2515 			md.md_vaddr = data_start;
2516 			md.md_size = data_end - data_start;
2517 			break;
2518 		case 2:
2519 			/* 3rd: kernel VM. */
2520 			va = prev->md_vaddr + prev->md_size;
2521 			/* Find start of next chunk (from va). */
2522 			while (va < virtual_end) {
2523 				/* Don't dump the buffer cache. */
2524 				if (va >= kmi.buffer_sva &&
2525 				    va < kmi.buffer_eva) {
2526 					va = kmi.buffer_eva;
2527 					continue;
2528 				}
2529 				pte = pte_find(mmu, kernel_pmap, va);
2530 				if (pte != NULL && PTE_ISVALID(pte))
2531 					break;
2532 				va += PAGE_SIZE;
2533 			}
2534 			if (va < virtual_end) {
2535 				md.md_vaddr = va;
2536 				va += PAGE_SIZE;
2537 				/* Find last page in chunk. */
2538 				while (va < virtual_end) {
2539 					/* Don't run into the buffer cache. */
2540 					if (va == kmi.buffer_sva)
2541 						break;
2542 					pte = pte_find(mmu, kernel_pmap, va);
2543 					if (pte == NULL || !PTE_ISVALID(pte))
2544 						break;
2545 					va += PAGE_SIZE;
2546 				}
2547 				md.md_size = va - md.md_vaddr;
2548 				break;
2549 			}
2550 			md.md_index = 3;
2551 			/* FALLTHROUGH */
2552 		default:
2553 			return (NULL);
2554 		}
2555 	} else { /* minidumps */
2556 		mem_regions(&physmem_regions, &physmem_regions_sz,
2557 		    &availmem_regions, &availmem_regions_sz);
2558 
2559 		if (prev == NULL) {
2560 			/* first physical chunk. */
2561 			md.md_paddr = physmem_regions[0].mr_start;
2562 			md.md_size = physmem_regions[0].mr_size;
2563 			md.md_vaddr = ~0UL;
2564 			md.md_index = 1;
2565 		} else if (md.md_index < physmem_regions_sz) {
2566 			md.md_paddr = physmem_regions[md.md_index].mr_start;
2567 			md.md_size = physmem_regions[md.md_index].mr_size;
2568 			md.md_vaddr = ~0UL;
2569 			md.md_index++;
2570 		} else {
2571 			/* There's no next physical chunk. */
2572 			return (NULL);
2573 		}
2574 	}
2575 
2576 	return (&md);
2577 }
2578 
2579 /*
2580  * Map a set of physical memory pages into the kernel virtual address space.
2581  * Return a pointer to where it is mapped. This routine is intended to be used
2582  * for mapping device memory, NOT real memory.
2583  */
2584 static void *
2585 mmu_booke_mapdev(mmu_t mmu, vm_offset_t pa, vm_size_t size)
2586 {
2587 	void *res;
2588 	uintptr_t va;
2589 	vm_size_t sz;
2590 
2591 	va = (pa >= 0x80000000) ? pa : (0xe2000000 + pa);
2592 	res = (void *)va;
2593 
2594 	do {
2595 		sz = 1 << (ilog2(size) & ~1);
2596 		if (bootverbose)
2597 			printf("Wiring VA=%x to PA=%x (size=%x), "
2598 			    "using TLB1[%d]\n", va, pa, sz, tlb1_idx);
2599 		tlb1_set_entry(va, pa, sz, _TLB_ENTRY_IO);
2600 		size -= sz;
2601 		pa += sz;
2602 		va += sz;
2603 	} while (size > 0);
2604 
2605 	return (res);
2606 }
2607 
2608 /*
2609  * 'Unmap' a range mapped by mmu_booke_mapdev().
2610  */
2611 static void
2612 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2613 {
2614 	vm_offset_t base, offset;
2615 
2616 	/*
2617 	 * Unmap only if this is inside kernel virtual space.
2618 	 */
2619 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2620 		base = trunc_page(va);
2621 		offset = va & PAGE_MASK;
2622 		size = roundup(offset + size, PAGE_SIZE);
2623 		kmem_free(kernel_map, base, size);
2624 	}
2625 }
2626 
2627 /*
2628  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2629  * specified pmap. This eliminates the blast of soft faults on process startup
2630  * and immediately after an mmap.
2631  */
2632 static void
2633 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2634     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2635 {
2636 
2637 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
2638 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2639 	    ("mmu_booke_object_init_pt: non-device object"));
2640 }
2641 
2642 /*
2643  * Perform the pmap work for mincore.
2644  */
2645 static int
2646 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2647     vm_paddr_t *locked_pa)
2648 {
2649 
2650 	TODO;
2651 	return (0);
2652 }
2653 
2654 /**************************************************************************/
2655 /* TID handling */
2656 /**************************************************************************/
2657 
2658 /*
2659  * Allocate a TID. If necessary, steal one from someone else.
2660  * The new TID is flushed from the TLB before returning.
2661  */
2662 static tlbtid_t
2663 tid_alloc(pmap_t pmap)
2664 {
2665 	tlbtid_t tid;
2666 	int thiscpu;
2667 
2668 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2669 
2670 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2671 
2672 	thiscpu = PCPU_GET(cpuid);
2673 
2674 	tid = PCPU_GET(tid_next);
2675 	if (tid > TID_MAX)
2676 		tid = TID_MIN;
2677 	PCPU_SET(tid_next, tid + 1);
2678 
2679 	/* If we are stealing TID then clear the relevant pmap's field */
2680 	if (tidbusy[thiscpu][tid] != NULL) {
2681 
2682 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2683 
2684 		tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2685 
2686 		/* Flush all entries from TLB0 matching this TID. */
2687 		tid_flush(tid);
2688 	}
2689 
2690 	tidbusy[thiscpu][tid] = pmap;
2691 	pmap->pm_tid[thiscpu] = tid;
2692 	__asm __volatile("msync; isync");
2693 
2694 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2695 	    PCPU_GET(tid_next));
2696 
2697 	return (tid);
2698 }
2699 
2700 /**************************************************************************/
2701 /* TLB0 handling */
2702 /**************************************************************************/
2703 
2704 static void
2705 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
2706     uint32_t mas7)
2707 {
2708 	int as;
2709 	char desc[3];
2710 	tlbtid_t tid;
2711 	vm_size_t size;
2712 	unsigned int tsize;
2713 
2714 	desc[2] = '\0';
2715 	if (mas1 & MAS1_VALID)
2716 		desc[0] = 'V';
2717 	else
2718 		desc[0] = ' ';
2719 
2720 	if (mas1 & MAS1_IPROT)
2721 		desc[1] = 'P';
2722 	else
2723 		desc[1] = ' ';
2724 
2725 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
2726 	tid = MAS1_GETTID(mas1);
2727 
2728 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2729 	size = 0;
2730 	if (tsize)
2731 		size = tsize2size(tsize);
2732 
2733 	debugf("%3d: (%s) [AS=%d] "
2734 	    "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2735 	    "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2736 	    i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2737 }
2738 
2739 /* Convert TLB0 va and way number to tlb0[] table index. */
2740 static inline unsigned int
2741 tlb0_tableidx(vm_offset_t va, unsigned int way)
2742 {
2743 	unsigned int idx;
2744 
2745 	idx = (way * TLB0_ENTRIES_PER_WAY);
2746 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2747 	return (idx);
2748 }
2749 
2750 /*
2751  * Invalidate TLB0 entry.
2752  */
2753 static inline void
2754 tlb0_flush_entry(vm_offset_t va)
2755 {
2756 
2757 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2758 
2759 	mtx_assert(&tlbivax_mutex, MA_OWNED);
2760 
2761 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2762 	__asm __volatile("isync; msync");
2763 	__asm __volatile("tlbsync; msync");
2764 
2765 	CTR1(KTR_PMAP, "%s: e", __func__);
2766 }
2767 
2768 /* Print out contents of the MAS registers for each TLB0 entry */
2769 void
2770 tlb0_print_tlbentries(void)
2771 {
2772 	uint32_t mas0, mas1, mas2, mas3, mas7;
2773 	int entryidx, way, idx;
2774 
2775 	debugf("TLB0 entries:\n");
2776 	for (way = 0; way < TLB0_WAYS; way ++)
2777 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2778 
2779 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2780 			mtspr(SPR_MAS0, mas0);
2781 			__asm __volatile("isync");
2782 
2783 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
2784 			mtspr(SPR_MAS2, mas2);
2785 
2786 			__asm __volatile("isync; tlbre");
2787 
2788 			mas1 = mfspr(SPR_MAS1);
2789 			mas2 = mfspr(SPR_MAS2);
2790 			mas3 = mfspr(SPR_MAS3);
2791 			mas7 = mfspr(SPR_MAS7);
2792 
2793 			idx = tlb0_tableidx(mas2, way);
2794 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2795 		}
2796 }
2797 
2798 /**************************************************************************/
2799 /* TLB1 handling */
2800 /**************************************************************************/
2801 
2802 /*
2803  * TLB1 mapping notes:
2804  *
2805  * TLB1[0]	CCSRBAR
2806  * TLB1[1]	Kernel text and data.
2807  * TLB1[2-15]	Additional kernel text and data mappings (if required), PCI
2808  *		windows, other devices mappings.
2809  */
2810 
2811 /*
2812  * Write given entry to TLB1 hardware.
2813  * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2814  */
2815 static void
2816 tlb1_write_entry(unsigned int idx)
2817 {
2818 	uint32_t mas0, mas7;
2819 
2820 	//debugf("tlb1_write_entry: s\n");
2821 
2822 	/* Clear high order RPN bits */
2823 	mas7 = 0;
2824 
2825 	/* Select entry */
2826 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2827 	//debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
2828 
2829 	mtspr(SPR_MAS0, mas0);
2830 	__asm __volatile("isync");
2831 	mtspr(SPR_MAS1, tlb1[idx].mas1);
2832 	__asm __volatile("isync");
2833 	mtspr(SPR_MAS2, tlb1[idx].mas2);
2834 	__asm __volatile("isync");
2835 	mtspr(SPR_MAS3, tlb1[idx].mas3);
2836 	__asm __volatile("isync");
2837 	mtspr(SPR_MAS7, mas7);
2838 	__asm __volatile("isync; tlbwe; isync; msync");
2839 
2840 	//debugf("tlb1_write_entry: e\n");
2841 }
2842 
2843 /*
2844  * Return the largest uint value log such that 2^log <= num.
2845  */
2846 static unsigned int
2847 ilog2(unsigned int num)
2848 {
2849 	int lz;
2850 
2851 	__asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
2852 	return (31 - lz);
2853 }
2854 
2855 /*
2856  * Convert TLB TSIZE value to mapped region size.
2857  */
2858 static vm_size_t
2859 tsize2size(unsigned int tsize)
2860 {
2861 
2862 	/*
2863 	 * size = 4^tsize KB
2864 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
2865 	 */
2866 
2867 	return ((1 << (2 * tsize)) * 1024);
2868 }
2869 
2870 /*
2871  * Convert region size (must be power of 4) to TLB TSIZE value.
2872  */
2873 static unsigned int
2874 size2tsize(vm_size_t size)
2875 {
2876 
2877 	return (ilog2(size) / 2 - 5);
2878 }
2879 
2880 /*
2881  * Register permanent kernel mapping in TLB1.
2882  *
2883  * Entries are created starting from index 0 (current free entry is
2884  * kept in tlb1_idx) and are not supposed to be invalidated.
2885  */
2886 static int
2887 tlb1_set_entry(vm_offset_t va, vm_offset_t pa, vm_size_t size,
2888     uint32_t flags)
2889 {
2890 	uint32_t ts, tid;
2891 	int tsize;
2892 
2893 	if (tlb1_idx >= TLB1_ENTRIES) {
2894 		printf("tlb1_set_entry: TLB1 full!\n");
2895 		return (-1);
2896 	}
2897 
2898 	/* Convert size to TSIZE */
2899 	tsize = size2tsize(size);
2900 
2901 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
2902 	/* XXX TS is hard coded to 0 for now as we only use single address space */
2903 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
2904 
2905 	/* XXX LOCK tlb1[] */
2906 
2907 	tlb1[tlb1_idx].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
2908 	tlb1[tlb1_idx].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
2909 	tlb1[tlb1_idx].mas2 = (va & MAS2_EPN_MASK) | flags;
2910 
2911 	/* Set supervisor RWX permission bits */
2912 	tlb1[tlb1_idx].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
2913 
2914 	tlb1_write_entry(tlb1_idx++);
2915 
2916 	/* XXX UNLOCK tlb1[] */
2917 
2918 	/*
2919 	 * XXX in general TLB1 updates should be propagated between CPUs,
2920 	 * since current design assumes to have the same TLB1 set-up on all
2921 	 * cores.
2922 	 */
2923 	return (0);
2924 }
2925 
2926 static int
2927 tlb1_entry_size_cmp(const void *a, const void *b)
2928 {
2929 	const vm_size_t *sza;
2930 	const vm_size_t *szb;
2931 
2932 	sza = a;
2933 	szb = b;
2934 	if (*sza > *szb)
2935 		return (-1);
2936 	else if (*sza < *szb)
2937 		return (1);
2938 	else
2939 		return (0);
2940 }
2941 
2942 /*
2943  * Map in contiguous RAM region into the TLB1 using maximum of
2944  * KERNEL_REGION_MAX_TLB_ENTRIES entries.
2945  *
2946  * If necessary round up last entry size and return total size
2947  * used by all allocated entries.
2948  */
2949 vm_size_t
2950 tlb1_mapin_region(vm_offset_t va, vm_offset_t pa, vm_size_t size)
2951 {
2952 	vm_size_t entry_size[KERNEL_REGION_MAX_TLB_ENTRIES];
2953 	vm_size_t mapped_size, sz, esz;
2954 	unsigned int log;
2955 	int i;
2956 
2957 	CTR4(KTR_PMAP, "%s: region size = 0x%08x va = 0x%08x pa = 0x%08x",
2958 	    __func__, size, va, pa);
2959 
2960 	mapped_size = 0;
2961 	sz = size;
2962 	memset(entry_size, 0, sizeof(entry_size));
2963 
2964 	/* Calculate entry sizes. */
2965 	for (i = 0; i < KERNEL_REGION_MAX_TLB_ENTRIES && sz > 0; i++) {
2966 
2967 		/* Largest region that is power of 4 and fits within size */
2968 		log = ilog2(sz) / 2;
2969 		esz = 1 << (2 * log);
2970 
2971 		/* If this is last entry cover remaining size. */
2972 		if (i ==  KERNEL_REGION_MAX_TLB_ENTRIES - 1) {
2973 			while (esz < sz)
2974 				esz = esz << 2;
2975 		}
2976 
2977 		entry_size[i] = esz;
2978 		mapped_size += esz;
2979 		if (esz < sz)
2980 			sz -= esz;
2981 		else
2982 			sz = 0;
2983 	}
2984 
2985 	/* Sort entry sizes, required to get proper entry address alignment. */
2986 	qsort(entry_size, KERNEL_REGION_MAX_TLB_ENTRIES,
2987 	    sizeof(vm_size_t), tlb1_entry_size_cmp);
2988 
2989 	/* Load TLB1 entries. */
2990 	for (i = 0; i < KERNEL_REGION_MAX_TLB_ENTRIES; i++) {
2991 		esz = entry_size[i];
2992 		if (!esz)
2993 			break;
2994 
2995 		CTR5(KTR_PMAP, "%s: entry %d: sz  = 0x%08x (va = 0x%08x "
2996 		    "pa = 0x%08x)", __func__, tlb1_idx, esz, va, pa);
2997 
2998 		tlb1_set_entry(va, pa, esz, _TLB_ENTRY_MEM);
2999 
3000 		va += esz;
3001 		pa += esz;
3002 	}
3003 
3004 	CTR3(KTR_PMAP, "%s: mapped size 0x%08x (wasted space 0x%08x)",
3005 	    __func__, mapped_size, mapped_size - size);
3006 
3007 	return (mapped_size);
3008 }
3009 
3010 /*
3011  * TLB1 initialization routine, to be called after the very first
3012  * assembler level setup done in locore.S.
3013  */
3014 void
3015 tlb1_init(vm_offset_t ccsrbar)
3016 {
3017 	uint32_t mas0;
3018 
3019 	/* TLB1[0] is used to map the kernel. Save that entry. */
3020 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
3021 	mtspr(SPR_MAS0, mas0);
3022 	__asm __volatile("isync; tlbre");
3023 
3024 	tlb1[0].mas1 = mfspr(SPR_MAS1);
3025 	tlb1[0].mas2 = mfspr(SPR_MAS2);
3026 	tlb1[0].mas3 = mfspr(SPR_MAS3);
3027 
3028 	/* Map in CCSRBAR in TLB1[1] */
3029 	tlb1_idx = 1;
3030 	tlb1_set_entry(CCSRBAR_VA, ccsrbar, CCSRBAR_SIZE, _TLB_ENTRY_IO);
3031 
3032 	/* Setup TLB miss defaults */
3033 	set_mas4_defaults();
3034 }
3035 
3036 /*
3037  * Setup MAS4 defaults.
3038  * These values are loaded to MAS0-2 on a TLB miss.
3039  */
3040 static void
3041 set_mas4_defaults(void)
3042 {
3043 	uint32_t mas4;
3044 
3045 	/* Defaults: TLB0, PID0, TSIZED=4K */
3046 	mas4 = MAS4_TLBSELD0;
3047 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3048 #ifdef SMP
3049 	mas4 |= MAS4_MD;
3050 #endif
3051 	mtspr(SPR_MAS4, mas4);
3052 	__asm __volatile("isync");
3053 }
3054 
3055 /*
3056  * Print out contents of the MAS registers for each TLB1 entry
3057  */
3058 void
3059 tlb1_print_tlbentries(void)
3060 {
3061 	uint32_t mas0, mas1, mas2, mas3, mas7;
3062 	int i;
3063 
3064 	debugf("TLB1 entries:\n");
3065 	for (i = 0; i < TLB1_ENTRIES; i++) {
3066 
3067 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3068 		mtspr(SPR_MAS0, mas0);
3069 
3070 		__asm __volatile("isync; tlbre");
3071 
3072 		mas1 = mfspr(SPR_MAS1);
3073 		mas2 = mfspr(SPR_MAS2);
3074 		mas3 = mfspr(SPR_MAS3);
3075 		mas7 = mfspr(SPR_MAS7);
3076 
3077 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3078 	}
3079 }
3080 
3081 /*
3082  * Print out contents of the in-ram tlb1 table.
3083  */
3084 void
3085 tlb1_print_entries(void)
3086 {
3087 	int i;
3088 
3089 	debugf("tlb1[] table entries:\n");
3090 	for (i = 0; i < TLB1_ENTRIES; i++)
3091 		tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0);
3092 }
3093 
3094 /*
3095  * Return 0 if the physical IO range is encompassed by one of the
3096  * the TLB1 entries, otherwise return related error code.
3097  */
3098 static int
3099 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3100 {
3101 	uint32_t prot;
3102 	vm_paddr_t pa_start;
3103 	vm_paddr_t pa_end;
3104 	unsigned int entry_tsize;
3105 	vm_size_t entry_size;
3106 
3107 	*va = (vm_offset_t)NULL;
3108 
3109 	/* Skip invalid entries */
3110 	if (!(tlb1[i].mas1 & MAS1_VALID))
3111 		return (EINVAL);
3112 
3113 	/*
3114 	 * The entry must be cache-inhibited, guarded, and r/w
3115 	 * so it can function as an i/o page
3116 	 */
3117 	prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3118 	if (prot != (MAS2_I | MAS2_G))
3119 		return (EPERM);
3120 
3121 	prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3122 	if (prot != (MAS3_SR | MAS3_SW))
3123 		return (EPERM);
3124 
3125 	/* The address should be within the entry range. */
3126 	entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3127 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3128 
3129 	entry_size = tsize2size(entry_tsize);
3130 	pa_start = tlb1[i].mas3 & MAS3_RPN;
3131 	pa_end = pa_start + entry_size - 1;
3132 
3133 	if ((pa < pa_start) || ((pa + size) > pa_end))
3134 		return (ERANGE);
3135 
3136 	/* Return virtual address of this mapping. */
3137 	*va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3138 	return (0);
3139 }
3140